Tag Archives: opinion polls

The Colossal Polling Errors of 1948 and 2016

1948: Dewey Defeats Truman. 2016: Clinton Will Defeat Trump. Both headlines ate crow.

Pollsters incorporate multiple factors when painting a picture of public opinion. (Try not spit in my face when you say that to me in person.) They consider race, gender, geographic location, religious affiliation, community setting, etc. They also need to consider which voters are more likely to show up. The most common method is by phone. Pollsters will hire people to call a bunch of numbers hoping that people will answer. At least a few hundred will. A lot of people don’t answer. I’ve been called by polling agencies a few times. I’ve answered and responded to questions twice.

What is the point of all this? Some of it is intended to feel the pulse of the nation — to see what Americans really want. Of course, a lot of it is to fuel the horse race that is an election. The media cites polls left and right. Some of it ends up influencing people’s votes. People are biased toward winners. FiveThirtyEight rose to prominence as being a polling agency. In 2008, they said Indiana and North Carolina would go for McCain, if I remember correctly. They went for Obama instead. In 2012, their presidential forecast was spot on. A lot of this is that they adjusted their model. They realized that the polls didn’t tell the whole story. There are other factors like the economy and the fact that pollsters can’t reach certain people by phone. They’ve had difficulty reaching Latino voters and people who may not have a phone at all.

In 2016, the polls could not have been much more wrong. FiveThirtyEight had a projection of about a 71.4% chance of Hillary Clinton winning, one of the more conservative estimates. People were betting money on PredictIt.com that Hillary would win, giving her better odds. The LA Times had one of the more accurate predictions. To have the final outcome be off by 3% or even 4% is not uncommon. That’s the margin of error for a lot of these polls. The polls showed Hillary Clinton leading by 4%. She is currently winning the popular vote by only 2%. However, even that lead doesn’t win the presidency. The “how do I explain this to a five-year-old child” system of the Electoral College chooses the victor. FiveThirtyEight was predicting a comfortable win in the electoral vote, maybe even a landslide. They predicted that Clinton would win all the Obama 2008 states except for Iowa, Indiana, and Ohio – 322 electoral votes in total.

Not only were the polls wrong in some of these states, they also had no idea that Donald Trump would steal states from the Democrats. Results in Michigan, Wisconsin, and Pennsylvania showed that Hillary Clinton didn’t just have an Iowa and Ohio problem, she had a Rust Belt problem. The last time a presidential candidate only won two states in the Midwest and won the presidency was 1884. How could the polls be so wrong in states that hadn’t voted for a Republican presidential candidate since the 1980s? One theory is the “shy Trump” voter avoided expressing support for Trump for fear of judgment by the pollster, but this is crap because this wasn’t a factor in the primaries. What is more likely is that Trump supporters were less likely to answer the phone. Knowing Trump’s impatience, you can see why his supporters would be reluctant to talk to some stranger on the phone for 10-15 minutes. Another issue is that pollsters vastly underestimated the number of working-class whites turning out to vote. Trump ended up winning with 306 electoral votes.

Let’s compare that to 1948. All the polls were saying that the suave Governor Thomas Dewey of New York would defeat the fiery incumbent president Harry Truman. Truman’s own staff members left for other jobs because they didn’t even think he could win. Gallup had Dewey winning the popular vote by 5%. He ended up losing it by 4.5%. In terms of the popular vote, that is clearly a greater upset than the 2016 election. I don’t have much to compare the electoral vote results of 1948 and 2016 as state polling for 1948 is hard to find.

A similarity between both elections is that a swing of 1.25% in three states were all that were needed to change the results. In 1948, those states were Ohio, Illinois, and California.  In 2016, those states were Wisconsin, Michigan, and Pennsylvania (alternate being Florida). Both elections would result in one party controlling the presidency and Congress. Additionally, polls in both elections showed a tightening race. In 1948, Dewey lost a double-digit lead, but experts still wouldn’t hedge their bets on Dewey. In 2016, FiveThirtyEight showed Hillary dropping from a 7% advantage to a 3.6% advantage.

FiveThirtyEight appeared to hedge its bets more than other predictions did and much more than most firms did in 1948. That still doesn’t excuse the fact that the polls did not capture the whole picture of what was going on the Rust Belt. The site even argued that 2016 probably wouldn’t be a repeat of 1948, and that Trump shouldn’t “bank on a massive polling error.” Earlier this year, polls showed Hillary Clinton winning Michigan’s primary over Bernie Sanders. Bernie ended up taking the state. So, a polling error in the Michigan is not without precedent. It’s almost as if we were supposed to see this coming, but we didn’t. Life goes on. We live and learn.

(Seriously, 2016 though. If I could go back in time, I would put money on DiCaprio, the Cavs, the Cubs, and Trump. How else could I become a millionaire?)

Advertisements